"How I can test it?" perhaps you don't even need one? build a prototype circuit and test the current arrangement, and see what the noise is like. For a supposedly non-moving reference, see what the ADC is reading like. If it stays steady, and noise is less than 1 LSB, then don't do anything.
If the noise is still sort of low, but you can see it oscillates around a particular value, try software averaging (take 10 samples, divide by 10, use the result as the 'reading').
If the noise is pretty bad, and averaging doesn't really help, you can then attempt the LPF.
The LPF would be best just before the ADC, so that it can filter out any strangeness emanating from the buffer op-amp as well as the original signal itself.
Be sure to check the expected input impedance of the ADC though, and ensure the output of the LPF still qualifies for this range, else the ADC may give incorrect values (may be off in scale, or slower than the input signal more than necessary, due to the sampling capacitors not getting charged quick enough, or too fast)